A Comparative Analysis of Discourse Marker Usage in Human and Machine Translation: Forms, Functions, and Distribution
Year: 2025
Type: article
Abstract: This research presents an in-depth comparative study of the use of discourse markers (DM) in human translation (HT) and machine translation (MT), their features, functions, and frequency across various discourse categories. Based on a mixed-methods approach, this study compares a parallel translation corpus to identify patterns of DM usage, evaluate their effect on coherence and fluency, and ascertain the extent to which MT systems replicate human-like discourse structure. Quantitative analysis reveals that HT employs a wider variety of DMs, distributing them strategically among functional categories such as contrastive, elaborative, inferential, and temporal markers. MT, by contrast, is characterized by an overreliance on a limited set of DMs, particularly inferential markers such as "so," and a lack of utilization of contrastive and reason markers essential for logical cohesion. This imbalance results in MT outputs that are rigid, redundant, or pragmatically flawed. Qualitative findings highlight how HT conveys context sensitivity and pragmatic appropriateness in selecting DM, bringing about smoothness and naturalness of discourse. In contrast, algorithmically constrained MT systems tend to handle DM insertion inappropriately, leading to instances of incoherence, abruptness of change, and loss of subtle meaning. The study also identifies that human translators convey implicit discourse relations, which MT struggles to express, instead falling back on explicit and superficial DM use at the cost of text subtlety. The findings have significant implications for both translation studies and computational linguistics. Pedagogically, the study emphasizes the need for translator training programs to incorporate an enhanced understanding of DM functionality and cross-linguistic variation. To enable MT development, the study suggests integrating discourse-aware training methods, which enhance the ability of neural networks to recognize contextual dependencies and dynamically optimize DM placement. The study further suggests that AI-supported post-editing methods can be extremely effective in making up for MT's discourse-level weaknesses, promoting coherence and fluency in machine translations.
Authors: Hafedh Hussein Al-Jassani, Doaa, Sadegh Barzani, Elahe, Mohsin Matter Al-Mawla, Fida, Karimi, Fatemeh
Institutions: Islamic Azad University, Isfahan, University of Wasit
Language: English
Subfield: Language and Linguistics
Field: Arts and Humanities
Domain: Social Sciences
Open Access status: diamond
OpenAlex ID: W6940598117
FWCI:
Cites:
Cited by:
Related to: